It has been 3 years since I posted anything on X, formerly known as twitter. Time for me to post something to make sure that my xeet embed still works.
@lefthanddraft It's clear that Quen 2.5 70b fine tuned on 800000 samples from deepseek-r1 would reason itself into doing things that would worry alignment researchers.
@theshawwn This one probably isn't fair. While GPT-4 doesn't necessarily give you the response directly. It gives you the code to get your response (second picture). In bard, I added a \g separately when it only matched the first character. GPT-4 intuited replacement of all characters.
@theshawwn I tried asking for some text changes in text from prev. responses. It only superficially followed directions. An example is asking it to replace every instance of I with 5 in the last response. There were a few replacements. Is were left in text. Random 5 were also strewn about.
@theshawwn I think that you can use the Facebook VOD in: https://t.co/a55HmNFKRb
* Go to https://t.co/0jAM98PsWY
* In the Facebook URL link enter: https://t.co/U7hFPWJfbo
--
Download video
Reading @MetaAI's Segment-Anything, and I believe today is one of the "GPT-3 moments" in computer vision. It has learned the *general* concept of what an "object" is, even for unknown objects, unfamiliar scenes (e.g. underwater & cell microscopy), and ambiguous cases.
I still can't believe both the model and data (11M images, 1B masks) are OPEN-sourced. Wow.๐ฎ
What's the secret sauce? Just follow the foundation model mindset:
1. A very simple but scalable architecture that takes multimodal prompts: text, key points, bounding boxes.
2. Intuitive human annotation pipeline that goes hand-in-hand with the model design.
3. A data flywheel that allows the model to bootstrap itself to tons of unlabeled images.
IMHO, Segment-Anything has done everything right.
This paper was written by @VinAI_Research
Authors: Thanh Van Le, Hao Phung, Thuan Hoang Nguyen, Quan Dao, Ngoc Tran and Anh Tran.
Six people and nobody thought to discuss ethics of poisoning.
People want a technical method to mark image intent, preventing AI from generating their faces. Current research seems malicious & unethical, ignoring liability of purposely damaging models.
https://t.co/KVFtEzqnmI
Ethics of poisoning not even discussed in https://t.co/stJEjI7yB8
While playing around with hooking up GPT-4 to the Internet, I asked it about myselfโฆ and had an absolute WTF moment before realizing that I wrote a very special secret message to Bing when Sydney came out and then forgot all about it. Indirect prompt injection is gonna be WILD
@GLACIULCINDER Liked this tweet because you shared it. If I saw your phone and liked this tweet as a result of seeing it on your phoneโฆ. That would be weird. Whoโd do that?!? Keep your screen covered friend. Keep your windows covered friend.
#Klingon seems to be trending. Don't forget these lovely Klingons; From the 1996 FMV Game; "Star Trek: Klingon", directed by Jonathan Frakes @jonathansfrakes. Characters pictured: Martha Hackett, John Cothran, J.G. Hertzler, Joe Hulser and two others that I'm not sure.